AWS has published a reference architecture for securely sharing a single Amazon SageMaker HyperPod EKS cluster among multiple teams. The design combines several components: AWS IAM Identity Center handles authentication, while per-team SageMaker Domains and Kubernetes namespaces provide isolation between groups. HyperPod Task Governance is used to enforce fairness across competing workloads, and namespace-level cost allocation supports chargeback to individual teams.
Why it matters
The approach aims to let organizations consolidate GPU resources into a shared cluster rather than provisioning separate environments per team, while maintaining boundaries for security, workload fairness, and cost attribution.
Who should care
Teams running machine learning workloads on AWS that need to share GPU clusters across multiple groups, along with platform and infrastructure engineers responsible for isolation, governance, and cost allocation.